QMW–PH–95–?
hep-th/yymmnnn
The Theory of Induced
Representations in Field Theory
´
Jose M. Figueroa-O’Farrill
ABSTRACT
These are some notes on Mackey’s theory of induced representations. They are based on lectures at the ITP in Stony Brook in the Spring of 1987. They were intended as an exercise in the theory of homogeneous spaces (as principal bundles). They were never meant for distribution.
§1 Introduction
The theory of group representations is by no means a closed chapter in mathematics. Although for some large classes of groups all representations are more or less completely classified, this is not the case for all groups. The theory of induced representations is a method of obtaining representations of a topological group starting from a representation of a subgroup. The classic example and one of fundamental importance in physics is the Wigner construction of representations of the Poincar´e group. Later Mackey systematized this construction and made it applicable to a large class of groups.
The method of induced representations appears geometrically very natural when expressed in the context of (homogeneous) vector bundles over a coset manifold. In fact, the representation of a group G induced from a representation of a subgroup H will be decomposed as a “direct integral” indexed by the elements of the space of cosets G/H. The induced representation of G will be carried by the completion of a suitable subspace of the space of sections through a given vector bundle over G/H.
In §2 we review the basic notions about coset manifolds emphasizing their connections with principal fibre bundles. In doing so we find it instructive to review the basic concepts associated with fibre bundles. The reader familiar with this should only scan this section to familiarize him/herself with the notation. In §3 we give the construction of induced representations of G carried by the space of sections through a homogeneous vector bundle or equivalently by functions defined on the group subject to an equivariance property. In §4 we discuss the unitarity of the induced representation and introduce the concept of a multiplier representation. In §5 we restrict ourselves to a special class of Lie groups (containing the spacetime groups) for which the method of induced representations yields all irreducible unitary representations and we shall give a method by which to obtain them. The method will be a generalization of Wigner’s “little group” method. This usually gives us a representation in a different space than the one we would like to use and, moreover, covariance under H transformations is often not manifest. Hence in §6 we introduce the so-called covariant functions and present a method to relate induced representations on one coset — where the representations are irreducible — to another coset where the representations are more useful. This will be of fundamental importance since the constraints we must impose to guarantee irreducibility in the second coset are — in the case of the Poincar´e group — nothing but the free field equations. Finally in §7 we look at some familiar examples in detail.
The following is a list of references — by no means complete — from which these notes have evolved:
– 2 –
[BR] A. O. Barut and R. Ra¸czka, Theory of Group Representations and Applications.
[H1] R. Hermann, Lie Groups for Physicists. [H2] R. Hermann, Fourier Analysis on Groups and Partial Wave Analysis. [Ma] G. W. Mackey, Induced Representations of Groups and Quantum Mechanics.
[NO] U. H. Niederer and L. O’Raifeartaigh, Forschritte der Physik, 22, 111-158
(1974).
[NW] P. van Nieuwenhuizen and P. West, Chapter 10 of forthcoming book.
[We] S. Weinberg, Quantum Theory of Massless Particles.
§2 General facts about Coset Spaces
Given any group G and a subgroup H we have a natural equivalence relation between elements of G. We say that two elements g, g′ are equivalent (g ∼ g′) if g′ = g · h for some h ∈ H. This partitions G into equivalence classes called (left) H cosets and the collection of all such cosets is denoted G/H. There is a canonical map G → G/H sending each group element to the coset containing it. We write this map as g → [g]. If G is a topological group, then all cosets are homeomorphic (as subspaces of G) and, indeed, if G is a Lie group and H a Lie subgroup then G/H can be given a differentiable structure such that the canonical map is smooth and furthermore G → G/H is a fibration of G by H. We now review some basic notions about fibre bundles. We shall, in view of the applications we have in mind, work in the differentiable category. All our spaces will be differentiable manifolds and all our maps will be smooth.
A fibre bundle is a generalization of the cartesian product. It consists of two differentiable manifolds E, B called the total space and base respectively with a smooth surjection π: E → B such that every point p ∈ B has a neighbourhood U such that its preimage in E looks like a cartesian product. That
≃
is, there exists a diffeomorphism φU : U ×F−→π−1 (U) giving local coordinates to E, where F is a topological space called the typical fibre. Hence locally E looks like B×F although generally this is not the case in the large. If this is the case then the bundle is called trivial. We often omit mention of the fibre when
π
this is understood and refer to a fibre bundle simply as E−→B. Now cover B with a collection of trivializing neighbourhoods {Ui} and let p ∈ Ui ∩ Uj. Let φi and φj be the coordinate maps associated with Ui and Uj, respectively. Then the preimage in E of Ui ∩Uj can be given coordinates in two ways: either by φi or by φj. The change of coordinates corresponds to a reparametrization
– 3 – of the fibre. More explicitly the map φ−i 1 ◦ φj mapping (Ui ∩ Uj) × F to itself breaks up as id × gij where the gij: Ui ∩ Uj → DiffF are called the transition
functions.
Generally the transition functions take values in a more manageable subgroup of DiffF and we get different kinds of fibre bundles depending on this subgroup as well as on particular properties of F. For instance if F is a vector space (say, V) and the transition functions take values in GL(V) — i.e.the linear structure of V is preserved — then the bundle is called a vector bun- dle. Now let F be a Lie group (say, H) and let there be a smooth free action of H on E on the right (for definiteness) which preserves the fibred structure — i.e.the action of H does not move a point of E away from its fibre — or, symbolically, denoting the right action of h ∈ H by Rh, such that π ◦ Rh = π. In this case we say that E is fibred by H and the base B is diffeomorphic to the space of orbits E/H. The transition functions then take values in H and the whole structure is called a principal fibre bundle. In particular when E is a Lie group G and the action of H is just right multiplication then the base can be identified with the space of left cosets G/H.
A way to give coordinates to the coset space G/H is to use the coordinates already present in G. The way to do this is to choose smoothly a representative for each coset. That is, for every coset p ∈ G/H choose an element of σ(p) ∈ G such that [σ(p)] = p. In fibre bundle language such a map is called a section.
π
Given a bundle E−→B any (smooth) map σ: B → E such that π ◦ σ = id is called a (smooth) section. Whereas a vector bundle has many sections (e.g.the zero section assigning to every point in the base the zero vector in the fibre) the existence of a global section in a principal fibre bundle is equivalent to triviality. Indeed we show that given a global section we can construct a
∼
- diffeomorphism E B × H. Given any point e ∈ E map it to (π(e), h) where
- =
e = Rh (σ (π(e))). Because the action of H is transitive on each fibre and free such an h always exist and is unique. Moreover the map is smooth because the action of H and π are both smooth. Conversely a trivial principal fibre bundle admits a global section. In fact, a global section in this case is nothing but the graph of a smooth function B → H.
We shall, for calculational and notational convenience, work with a global section. This, although true for the case of contractible bases such as Minkowski spacetime, is not always possible. However if the base contains a subspace M whose complement is of measure zero (with respect to a suitable measure on the base) and such that the bundle over M is trivial we can safely (i.e.for the purposes we have in mind) assume that the bundle is trivial. We shall see later how this arises.
π
Given a principal fibre bundle E−→B with fibre H and a representation
– 4 –
D: H → GL(V) of H we can construct a vector bundle with fibre V called an associated vector bundle. The construction runs as follows. Consider the cartesian product E × V and define on it an equivalence relation as follows
(e, v) ∼ (Rh(e), D(h−1) · v) for all h ∈ H. That this is an equivalence relation follows from the fact that the action of H on E is a right action. The collection of equivalence classes is denoted E ×D V. This forms the total space of
π′
the associated vector bundle E ×D V−→B, where the projection is given by π′ ([(e, v)]) = π(e). It is straight forward to show that this is indeed a vector bundle. The local product structure is inherited from that of the original principal bundle and it is easy to see that the transition functions are just the representation of the original transition functions. In the particular case where E is a Lie group G then the associated vector bundle is called a homogeneous vector bundle. Unless otherwise stated we take V to be a complex vector space
throughout.
The sections through an associated vector bundle can be thought of as functions from the total space E of the principal bundle to the vector space V with the following equivariance condition. If f: E → V then f ◦ Rh =
D(h−1) ◦ f for all h ∈ H. In the context of induced representations these functions are known as Mackey functions. We shall denote by M the space of Mackey functions. It is possible to make M into a vector space by defining the relevant operations pointwise. That is, if f1, f2 ∈ M then we define f1 +f2 by (f1 + f2)(g) = f1(g) + f2(g) and similiarly for scalar multiplication. The same is true for the space of sections of any vector bundle. Let Γ denote the space of sections of the homogeneous vector bundle G ×D V−→G/H. We
now exhibit an isomorphism between M and Γ. It should be remarked that although the proof given here is for the case of a homogeneous vector bundle the result is true for an arbitrary associated vector bundle as the reader can easily verify.
Given a Mackey function f ∈ M we define a section as follows:
ψ(p) d=ef [(σ(p), f(σ(p)))]
∀p ∈ G/H,
(2.1) where p → σ(p) is a coset representative. It is easy to verify that it is well defined (i.e.independent of the coset representative), for if we choose another coset representative p → σ′(p) = σ(p) · h(p) for some h(p) ∈ H then
- [(
- )]
ψ′(p) d=ef σ′(p), f(σ′(p))
= [(σ(p) · h(p), f(σ(p) · h(p)))]
- [(
- )]
= σ(p) · h(p), D(h(p)−1) · f(σ(p)) = [(σ(p), f(σ(p)))]
– 5 –
=ψ(p) .
Furthermore it is clear that π ◦ σ = id, so ψ ∈ Γ.
Conversely given a section ψ ∈ Γ and a coset representative p → σ(p) we
e
construct a Mackey function ϕ as follows. Let g ∈ G and write it as σ(p) · h where p = [g]. Then we define
def
−1
e
ψ(g) = D(h ) · vp
,
(2.2)
e
where ψ(p) = [(σ(p), vp)]. Clearly ψ is equivariant by construction. Moreover it is again straight forward to show that the above defined Mackey function is independent of the choice of coset representative.
It is worth remarking that although we have assumed the existence of a global section G/H → G the above constructions are well defined even for non trivial bundles. The correspondence between M and Γ is defined locally and the proof just shown that it is independent of the choice of section can be reproduced mutatis mutandis to prove that the correspondence agrees in the intersection of trivializing neighbourhoods.
We remarked above that both spaces can be made into vector spaces by defining the relevant operations pointwise. In that case, the above correspondence can then be seen to be a vector space isomorphism.
In the particular case of a homogeneous vector bundle over G/H there is a natural action of G on the space of sections or equivalently on the Mackey
def
e
−1
functions. Let f be a Mackey function. Then we define Ug · f = f ◦ Lg where Lg represents left multiplication by g. That this forms a representation follows from the fact that left multiplication realizes the group and that the
e
space M of Mackey functions is a vector space. Then one verifies that Ug is
- e
- e
- e
linear with respect to this vector space structure and that Ug ·g = Ug · Ug .
- 1
- 2
- 1
- 2
This representation of G carried by M will induce a representation of G carried by the space Γ of sections and this will be what we call the representation of G induced by the representation D of H. It shall be topic of the ensuing sections, but before continuing with this topic it shall be convenient to develop some machinery to deal with calculations in coset spaces.
Consider the principal fibre bundle G−→G/H where the projection is the canonical map g → [g]. There is a natural action of G on the cosets G/H induced from left multiplication on G. That is, we define a map τ: G → Diff(G/H) such that τg ·[g′] = [g · g′]. It is trivial to verify that τg ·g = τg ·τg .
- 1
- 2
- 1
- 2
This action is obviously transitive so that any two points in G/H are connected via a G transformation. Now choose a coset representative (i.e.a section)
– 6 – p → σ(p). Then τg · p = [g · σ(p)] and, by definition, τg · p = [σ (τg · p)] which allows us to conclude that g · σ(p) ≡ σ(τg · p) mod H. We define a map
h: G/H × G → H by
- g · σ(p) = σ(τg · p) · h(p, g)
- ∀p ∈ G/H , g ∈ G .
(2.3)
Because H is the isotropy subgroup of the identity coset po — i.e.τh · po = po for all h ∈ H —
- h(po, h′) = Ad
- (h′)
∀h′ ∈ H .
(2.4)
σ(po)−1
Defining our coset representative in such a way that σ(po) is the identity — which we are always free to do performing, if necessary, a global H transformation — we find that h(po, h′) = h′ for all h′ ∈ H. Also from the definition we find that
h(p, g1 · g2) = h(τg · p, g1) · h(p, g2) .
(2.5)
2
This, together with the easily verifiable fact that h(po, σ(p)) = 1 yields h(po, σ(p)· h′) = h′.
We now consider in detail the following special case of which the Poincar´e group forms part. Let G be the semidirect product T o H. That is, T is a
normal subgroup of G and every element g ∈ G can be (uniquely) written as a product h · t where h ∈ H and t ∈ T. It then follows that as manifolds G/H and T are diffeomorphic. The correspondence is the following: Let g = h · t be any element of G, then [g] = [h · t] = [Adh(t) · h] = [Adh(t)] and we define the map [h · t] → Adh(t). This map is easily seen to be a diffeomorphism. A few straightforward calculations yield the following properties, where t ∈ T, h′ ∈ H and where we have chosen a coset representative in T ⊂ G:
h(p, t) = 1
(2.6) (2.7) (2.8) (2.9) σ(τt · p) = t · σ(p)
h(p, h′) = h′
- ′
- ′
σ(τh · p) = Adh (σ(p)) .
These equations will be very useful in the sequel.
– 7 –
§3 Induced Representations
We saw in the last section how there was a natural action of G on the vector space M of Mackey functions. We also saw that there was a vector space isomorphism between M and the vector space Γ of sections through the homogeneous vector bundle. This isomorphism will induce a representation of G on Γ which we now construct.
Fix a choice≃of coset representative and let us denote the corresponding
- e
- e
isomorphism Γ−→M by ψ → ψ. We denote by U the action of G on M. We induce an action U of G on Γ by demanding commutativity of the following diagram:
Γ −→ M
y
e
Ug
U
y
g
Γ −→ M
That is,
- e
- e
^
- Ug · ψ = Ug · ψ
- (∀g ∈ G) .
(3.1)
Explicitly,
- [(
- )]
)]
^
(Ug · ψ) (p) = σ(p), Ug · ψ(σ(p))
[(
- e
- e
=
σ(p), Ug · ψ(σ(p))
- [(
- )]
e
=
σ(p), ψ(g−1σ(p))
Recalling that g−1 · σ(p) = σ(τg · p) · h(p, g−1) and using the equivariance of
−1
the Mackey functions we find that ψ(g ·σ(p)) = D(h(p, g−1)−1)·ψ(σ(τg ·p)
−1
- e
- e
−1
and thus
- [(
- )]
(Ug · ψ) (p) = σ(p), D(h(p, g−1)−1) · ψσ(τg · p)
.
(3.2)
e
−1
Fixing a global coset representative (under the assumptions of §2) we may identify the fiber at p of G ×D V with V itself, although this identification,
depending on the choice of coset representative, is not canonical. Hence if ψ(p) = [(σ(p), vp)] then we may write ψ(p) = vp with a little abuse of notation.